JEPA-Anything: one predictive framework spanning vision, biology, weather and more
rbhar90 · x · 2026-09-19
A new arXiv paper (2609.20800) introduces JEPA-Anything, a domain-agnostic world-modeling framework built on Orthogonal Predictive Factorization (OPF), which splits JEPA latent targets into inspectable complementary factors instead of one monolithic target.
- Evaluated across 7 domains: vision, biology, clinical trajectories, control, molecular dynamics, physical fields, and weather
- Experiments include 10 matched dynamics tasks, forecasting of 1,000+ clinical events, and 100-step molecular rollouts across 4 systems
- Beats matched JEPA baselines on all 10 dynamics tasks, cuts single-intervention prediction error on Interventional Pong by 34.8%, and achieves the lowest molecular errors in all systems
- Code, checkpoints, and report fully open-sourced
More from Research
- Amazon: RAG must move beyond factual grounding to surface diverse opinions — _reachsumit · 2026-11-04
- ShiftRAG bypasses RAG's textual bottleneck by feeding retrieval signals as soft tokens — _reachsumit · 2026-11-04
- Meta's Atomic Creator Tokens beat Semantic IDs: +19.8pt Recall@15 and 2.7x lower latency — _reachsumit · 2026-11-04
- HiGenTrans transfers LLM knowledge into generative recommenders at input, representation and decision levels — _reachsumit · 2026-11-04
- SA-OPSD distills self-teacher knowledge at beam-search pruning boundaries for generative recommendation — _reachsumit · 2026-11-04
- MV-Thinker: Multi-View Thinking with Adaptive Generation for Generative Recommendation — _reachsumit · 2026-11-04